Journal · method
Session quality is not duration
Product meetings still treat average session length as a mood. Up is good. Down is a leak. That heuristic survives because it is easy to chart and because it sometimes correlates with curiosity. It is a weak proxy for whether the user accomplished anything.
In the Integrity Lab we start with two recordings that share almost the same duration. One is a grocery shopper who found a substitute, paid, and left. The other is a hospital booking user looping between specialty and doctor list because filters reset. Both sit near four minutes. Only one session is high quality.
What duration actually captures
Clock time captures attention spent inside your process, including attention you wasted. It also captures OS interruptions, incoming calls, and a child asking for a snack. Mobile sessions are full of pauses that have nothing to do with your UX.
Duration also hides completion that happens quickly. A returning user who knows the PromptPay flow can finish a bill pay in forty seconds. That session looks “thin” next to a first-time user who wandered. The returning user did the job. The wanderer may not have.
A better first question
Ask what job opened the session. Then ask whether that job moved. Completion is not always a purchase. It can be “check if the parcel passed customs,” “see if the doctor has a Saturday slot,” or “confirm the meter reading posted.” If those jobs finish, a short session is a compliment.
If the job did not move, duration becomes evidence of struggle or of idle confusion. That is when friction and recovery signals matter. A rage tap at minute three is not “engagement.” It is a person hitting a control that did not respond.
How we score without worshipping the clock
We still look at duration. We just demote it. In the four-signal sheet, duration appears only as a note under friction (“long stall on OCR”) or recovery (“returned after a 90-second hop”). It never gets its own point total.
Teams that insist on a single KPI sometimes build a “quality index” that still smuggles time back in through a weighted average. Resist that. If you need one number for an exec slide, use share of sessions with job completion, and put the limitation underneath in a readable font.
A limitation worth writing down
You often cannot observe the job with events alone. Family bookings, gift purchases, and “just looking” intent are under-instrumented. When that is true, say so. A score that pretends every session wanted to convert will punish browsers and flatter accidental checkouts.
Duration will always be on the dashboard. Treat it as weather, not as the harvest.